You will get a web scraper that keeps running and tells you the moment it breaks
Project details
Anyone can get a scraper written today. The hard part was never writing it - it is the Tuesday morning six weeks later when the site changed its HTML and your file is empty, or worse, half empty and nobody noticed.
I run a news pipeline that pulls from 70+ sources four times a day. Not long ago one source quietly started returning 130 characters instead of 3,000. Every dashboard stayed green. That is the failure mode I build against.
Single Source - one site, clean structured output, pagination handled.
Multi Source - up to 5 sites merged into one normalised dataset, deduplicated.
Monitored - up to 10 sites plus a watchdog that checks every run for empty results, shape changes and silent truncation, and alerts you the moment something drifts.
You get the full source code, a README you can actually follow, and a short screen recording of it running.
I only build scrapers for publicly accessible pages. I do not bypass logins, paywalls or CAPTCHAs, and I do not touch personal data.
My English is basic but clear. I reply within a few hours and I would rather over-explain than leave you guessing.
I run a news pipeline that pulls from 70+ sources four times a day. Not long ago one source quietly started returning 130 characters instead of 3,000. Every dashboard stayed green. That is the failure mode I build against.
Single Source - one site, clean structured output, pagination handled.
Multi Source - up to 5 sites merged into one normalised dataset, deduplicated.
Monitored - up to 10 sites plus a watchdog that checks every run for empty results, shape changes and silent truncation, and alerts you the moment something drifts.
You get the full source code, a README you can actually follow, and a short screen recording of it running.
I only build scrapers for publicly accessible pages. I do not bypass logins, paywalls or CAPTCHAs, and I do not touch personal data.
My English is basic but clear. I reply within a few hours and I would rather over-explain than leave you guessing.
Programming Languages
JavaScript, Python, TypeScriptCoding Expertise
Performance OptimizationWhat's included
| Service Tiers |
Starter
$90
|
Standard
$200
|
Advanced
$400
|
|---|---|---|---|
| Delivery Time | 5 days | 8 days | 12 days |
Number of Revisions | 1 | 2 | 2 |
Number of Sources Mined/Scraped | 1 | 5 | 10 |
Install Script | - | - | - |
Test Script | |||
Task Automation |
Frequently asked questions
About Duc
n8n & Self-Hosted AI Automation Engineer | Zero-API-Cost Pipelines
Vo Xu, Vietnam - 3:08 pm local time
For the past two years I have been running a fully self-hosted content and media automation stack, alone, end to end:
- 31 live n8n workflows orchestrating crawling, content generation, image generation, publishing and monitoring
- 123 scheduled jobs and services across 4 Linux/macOS servers, with a supervisor agent that reports system health to a chat channel every day
- A news pipeline that crawls 70+ active sources, deduplicates near-identical stories across outlets, runs quality and fact-check gates, generates original images, and publishes to a live website unattended, four times a day
- Self-hosted LLMs (14B + 35B) served locally behind an internal gateway with queueing, rate limiting and fallback, so the whole system runs at zero API cost
- Local image generation (FLUX) with a full asset pipeline: WebP/AVIF variants, CDN delivery, checksum deduplication, orphan cleanup
- A publisher that takes one video and distributes it to Facebook, YouTube and TikTok with platform-appropriate metadata
What I am good at, specifically:
- Cutting LLM API bills to near zero by moving workloads to self-hosted models, and routing only what genuinely needs a frontier model to a paid API
- Building content pipelines that do not produce garbage: deduplication, quality gates, fact-checking against the source material, and human approval steps where they matter
- Making automations observable. If a job dies at 3am you get a message, not a silent gap in your data
- Rescuing automations that worked in testing and fall over in production
Stack: n8n, Node.js, Python, PostgreSQL (+pgvector), Redis, Docker, systemd, local LLM serving on Apple Silicon and NVIDIA GPUs, Directus, Astro, Cloudflare (Pages, R2, Workers), Facebook Graph API, YouTube and TikTok APIs, Synology.
A note on how I work: I am based in Vietnam and I prefer asynchronous, written collaboration - written briefs, chat, screen recordings and clear documentation - over daily video calls. My spoken English is basic. If your project needs a lot of live conversation, I am probably not your best fit. If you want someone who ships and documents what he shipped, I am.
Tell me the process you are tired of doing by hand, and I will tell you honestly whether automating it is worth your money.
Steps for completing your project
After purchasing the project, send requirements so Duc can start the project.
Delivery time starts when Duc receives requirements from you.
Duc works on your project following the steps below.
Revisions may occur after the delivery date.
Check what is actually reachable
Some sites render everything server-side, some hide the data behind JavaScript, some rate-limit hard. I open your pages first and tell you which one yours is before I write a line of code.
Build it against the live site
I write the scraper and run it against the real pages, not a saved copy. Pagination, polite rate limiting and retries are handled from the start, not bolted on after it breaks.